Cursor 2, released on October 29, 2025, doubles down on being the best place to work with AI agents. The Composer frontier model now lives inside the editor, while a redesigned multi-agent canvas, agent-aware testing, and workflow upgrades keep developers shipping code faster than ever.
Cursor 2 runs in non-thinking mode by default, switching to deep reasoning only when complex refactors demand it.
Cursor 2 bundles Composer, multi-agent orchestration, browser debugging, and enterprise controls into a cohesive launch. Dive into the motion previews and guides below to experience every upgrade first-hand.
Manage every agent from a dedicated sidebar, orchestrate up to eight parallel runs on one prompt, and keep plans visible without juggling files.
Composer is Cursor’s first in-house agentic coding model, engineered to deliver low-latency turns that keep momentum high across large repositories.
Embed the browser directly in the editor so agents can select DOM elements, forward page context, and iterate on UI fixes without leaving Cursor 2.
Inspect every file the agent touched inside one diff viewer, with plans and context pills pinned next to the changes for quicker approvals.
Run agent-issued shell commands safely. Cursor 2 elevates the sandbox to GA, defaulting to secure macOS environments with limited workspace access.
Publish curated prompts, rules, and workflows from the Cursor dashboard so every teammate automates with the same playbook.
Dictate prompts, set custom trigger phrases, and let the agent execute tasks while you stay in flow or code hands-free.
Files, directories, and recent context now surface as inline pills, while the agent auto-gathers definitions, lint errors, and links for you.
Design plans with one model while another builds them, or let parallel agents draft multiple plans you can review side by side.
Faster Language Server Protocol startup, dynamic memory scaling for Python and TypeScript, and leak fixes make Cursor 2 feel instantly responsive.
Read performance notesA revamped harness boosts reliability across every model, with notable quality jumps for GPT-5 Codex inside cursor 2 projects.
Harness improvementsCloud agents now deliver 99.9% reliability, instant startup, and a refreshed UI on the horizon for sending runs beyond your laptop.
Cloud agent roadmapEnforce sandbox policies, distribute hooks from the dashboard, and audit every admin event to keep enterprise cursor 2 environments compliant.
Enterprise upgradesCursor 2 reimagines the VS Code-based editor as an AI-native workspace that elevates agents over individual files. Every workflow orbits the outcome you want—brief an agent, review its plan, and merge the best diff—so the path from idea to shipped commit is dramatically shorter.
Git worktrees or remote sandboxes isolate each agent run, while semantic search feeds Composer rich context across the entire repository. This “agents-first” architecture lets up to eight assistants iterate in parallel without clashing changes or missing long-range dependencies.
Launched on October 29, 2025 as a free upgrade, Cursor 2 formalized Cursor’s vision of being the best place to work with agents by bundling Composer, the new agent console, and quality-of-life tooling in one release.
Composer handles monorepos and legacy stacks with semantic recall, bridging gaps between services, CI pipelines, and historical diffs for richer reasoning.
Release notes, changelogs, and onboarding guides remain centralized at cursor.com, detailing every improvement from UI toggles back to the classic layout to agent plan management.
Composer is engineered for low-latency, agentic coding. Most edits, refactors, and reviews land in under 30 seconds—about 4× faster than similarly capable models—and its semantic-search training keeps context intact across sprawling codebases.
Engineers can launch up to eight parallel agents, compare results, and merge the strongest diff. A right-side sidebar tracks plans, outputs, and quick toggles back to the classic IDE view when needed.
A built-in browser environment lets agents run end-to-end tests, capture screenshots, and iterate on interface bugs without leaving the editor, graduating from beta to general availability in Cursor 2.
Hands-free input transforms spoken instructions into code edits. Voice Mode provides speech-to-text and custom trigger phrases so you can talk to the agent while staying in flow.
Memory usage drops dramatically, inference speeds climb, and review tooling delivers sharper diffs for deeper inspection. LSP features now load faster across languages and cloud agents boast 99.9% uptime with instant startup.
Reviewing agent work is now frictionless. Cursor 2 aggregates every file change in a unified diff viewer so you can inspect the full impact without hunting through folders. Context pills highlight referenced files, plans sit beside the diff, and switching back to the classic editor view is one click away.
The integrated browser tool graduates to general availability in Cursor 2, letting agents execute UI flows, select DOM elements, and keep iterating until the experience matches the prompt. This keeps the model inside a sandboxed environment while closing the loop between writing and validating code.
| Capability | Cursor 2 | Previous Cursor | Claude 4.5 Sonnet |
|---|---|---|---|
| Average task completion | Under 30 seconds with Composer | 1–2 minutes on complex edits | About 1.8 minutes |
| Token efficiency | ≈4.5× fewer tokens per diff | Moderate reduction with caching | High token footprint |
| Parallel agents | Agent-centric UI + git worktrees | Single agent focus | Limited concurrency |
| Browser debugging | Native browser tool with screenshots | Depends on external tools | No integrated browser |
| Pricing structure | Usage-based, Composer priced like GPT-5 | Credit-based plans | Higher per-token costs |
| Accessibility features | Voice-to-text with agent execution | Manual integrations required | No first-party tools |
Download the update from cursor.com/download and migrate your editor.
Set Composer as the default Auto mode model so everyday prompts enjoy low latency and richer context.
Launch parallel agents for tricky features. Git worktrees keep experiments isolated and conflicts in check.
Run UI tests, capture screenshots, and validate front-end fixes without swapping applications.
Convert spoken prompts into code changes, ideal for walkthroughs or when resting your hands.
Track completion time and token usage to optimize budgets and developer satisfaction.
Sync with your team on agent outputs, then publish shared prompts or Team Commands so every developer benefits from the strongest workflow.
Review changelog updates to capture new multi-agent tricks, sandbox improvements, and performance optimizations as they land.
Large teams praise Composer’s speed, crediting it with resolving long-standing bugs within minutes of parallel agent experimentation.
Some users note the agent-centric UI feels unfamiliar and warn that running multiple agents can increase token usage without guardrails.
Threads on Reddit highlight excitement for problem-solving speed while flagging issues like broken hot reloading in certain setups.
Agent-issued shell commands now run in secure sandboxes by default on macOS, inheriting the beta feature from v1.7 and granting read/write access only to your workspace while blocking unapproved network calls.
Team Commands let leaders ship curated prompts or rules across the organization through shareable deep links. Every teammate inherits the same presets inside Cursor 2 for consistent, policy-aligned automation.
Prompt UI enhancements surface referenced files as inline pills, and Composer can auto-gather definitions, recent changes, or lint errors. Combined with Voice Mode’s trigger phrases, developers juggle context hands-free.
Existing users receive Cursor 2 as a free update, keeping solo developers on the latest Composer model without extra charges.
Team plans move from credit-based to usage-based billing, aligning costs with real Composer consumption and repository size.
Although Composer costs match GPT-5 per token, efficiency gains mean equivalent tasks average 73% cheaper thanks to shorter responses.
Composer is tuned for low-latency prompts, completing most tasks in under 30 seconds by defaulting to non-thinking mode until deeper reasoning is required.
Cursor 2 lets you run agents in parallel and compare their diffs side by side, all while git worktrees keep branches isolated.
Yes. Semantic search and contextual retrieval allow Cursor 2 to navigate monorepos and legacy codebases without losing track of dependencies.
A native browser tool lets agents run client-side tests, capture screenshots, and debug UI issues directly in the editor.
Composer is priced like GPT-5 per token, but overall workflows cost less because Cursor 2 produces shorter, more targeted outputs.
Yes. Voice prompts travel through the same encrypted channels as text, keeping enterprise security intact.
A layout toggle returns the sidebar to its previous position so teams can transition gradually while testing multi-agent workflows.
Agents can trigger browser UI checks, run automated tests, and surface diffs, helping teams validate changes before merging.
The multi-agent hub accepts GPT-5, Claude 4.5 Sonnet, and future models, enabling head-to-head comparisons on the same task.
Visit cursor.com/changelog/2-0 for every enhancement, bug fix, and performance note tied to Cursor 2.